Magnetic resonance image segmentation using semi-automated software for quantification of knee articular cartilage---initial evaluation of a technique for paired scans

Accuracy and longitudinal reproducibility of quantitative femorotibial cartilage measures derived from automated U-Net-based segmentation of two different MRI contrasts: data from the osteoarthritis initiative healthy reference cohort

OBJECTIVE

To evaluate the agreement, accuracy, and longitudinal reproducibility of quantitative cartilage morphometry from 2D U-Net-based automated segmentations for 3T coronal fast low angle shot (corFLASH) and sagittal double echo at steady-state (sagDESS) MRI.

METHODS

2D U-Nets were trained using manual, quality-controlled femorotibial cartilage segmentations available for 92 Osteoarthritis Initiative healthy reference cohort participants from both corFLASH and sagDESS (n = 50/21/21 training/validation/test-set). Cartilage morphometry was computed from automated and manual segmentations for knees from the test-set. Agreement and accuracy were evaluated from baseline visits (dice similarity coefficient: DSC, correlation analysis, systematic offset). The longitudinal reproducibility was assessed from year-1 and -2 follow-up visits (root-mean-squared coefficient of variation, RMSCV%).

RESULTS

Automated segmentations showed high agreement (DSC 0.89-0.92) and high correlations (r ≥ 0.92) with manual ground truth for both corFLASH and sagDESS and only small systematic offsets (≤ 10.1%). The automated measurements showed a similar test-retest reproducibility over 1 year (RMSCV% 1.0-4.5%) as manual measurements (RMSCV% 0.5-2.5%).

DISCUSSION

The 2D U-Net-based automated segmentation method yielded high agreement compared with manual segmentation and also demonstrated high accuracy and longitudinal test-retest reproducibility for morphometric analysis of articular cartilage derived from it, using both corFLASH and sagDESS.

Simultaneous super-resolution and contrast synthesis of routine clinical magnetic resonance images of the knee for improving automatic segmentation of joint cartilage: data from the Osteoarthritis Initiative

PURPOSE

High resolution three-dimensional (3D) magnetic resonance (MR) images are well suited for automated cartilage segmentation in the human knee joint. However, volumetric scans such as 3D Double-Echo Steady-State (DESS) images are not routinely acquired in clinical practice which limits opportunities for reliable cartilage segmentation using (fully) automated algorithms. In this work, a method for generating synthetic 3D MR (syn3D-DESS) images with better contrast and higher spatial resolution from routine, low resolution, two-dimensional (2D) Turbo-Spin Echo (TSE) clinical knee scans is proposed.

METHODS

A UNet convolutional neural network is employed for synthesizing enhanced artificial MR images suitable for automated knee cartilage segmentation. Training of the model was performed on a large, publically available dataset from the OAI, consisting of 578 MR examinations of knee joints from 102 healthy individuals and patients with knee osteoarthritis.

RESULTS

The generated synthetic images have higher spatial resolution and better tissue contrast than the original 2D TSE, which allow high quality automated 3D segmentations of the cartilage. The proposed approach was evaluated on a separate set of MR images from 88 subjects with manual cartilage segmentations. It provided a significant improvement in automated segmentation of knee cartilages when using the syn3D-DESS images compared to the original 2D TSE images.

CONCLUSION

The proposed method can successfully synthesize 3D DESS images from 2D TSE images to provide images suitable for automated cartilage segmentation.

Quantification in Musculoskeletal Imaging Using Computational Analysis and Machine Learning: Segmentation and Radiomics

Although still limited in clinical practice, quantitative analysis is expected to increase the value of musculoskeletal (MSK) imaging. Segmentation aims at isolating the tissues and/or regions of interest in the image and is crucial to the extraction of quantitative features such as size, signal intensity, or image texture. These features may serve to support the diagnosis and monitoring of disease. Radiomics refers to the process of extracting large amounts of features from radiologic images and combining them with clinical, biological, genetic, or any other type of complementary data to build diagnostic, prognostic, or predictive models. The advent of machine learning offers promising prospects for automatic segmentation and integration of large amounts of data. We present commonly used segmentation methods and describe the radiomics pipeline, highlighting the challenges to overcome for adoption in clinical practice. We provide some examples of applications from the MSK literature.

Determination of knee cartilage volume and surface area in beagle dogs: a pilot study

BACKGROUND

The objective of this study was to determine the cartilage volume and surface area of male and female Beagle dog knees using 3D (3 dimensional) reconstructed MRI images.

METHODS

Six Beagle Dogs (Canis familiaris) (3 males and 3 females) of 10-18 months old and weighing between 7.2 and 17.1 kg underwent a MRI evaluation of both knees. The data acquired allowed a 3D reconstruction of the knee and measurement of the cartilage volume and surface area.

RESULTS

Mean knee cartilage volume (averaged over the right and left knees) of animals between 7.2 and 17.1 kg ranged from 319.7 to 647.3 mm3; while the mean knee cartilage surface area ranged from 427.14 to 757.2 mm2. There was evidence of both knee volume and surface area increasing linearly with animal bodyweight.

CONCLUSIONS

The cartilage volume and surface area of the Beagle dog appears to correlate significantly with body weight. This study provides a reference base for future studies investigating cartilage related pathology such as osteoarthritis.

Application of autofluorescence robotic histology for quantitative evaluation of the 3-dimensional morphology of murine articular cartilage

Murine models of osteoarthritis (OA) are increasingly important for understating pathogenesis and for testing new therapeutic approaches. Their translational potential is, however, limited by the reduced size of mouse limbs which requires a much higher resolution to evaluate their articular cartilage compared to clinical imaging tools. In experimental models, this tissue has been predominantly assessed by time-consuming histopathology using standardized semi-quantitative scoring systems. This study aimed to develop a novel imaging method for 3-dimensional (3D) histology of mouse articular cartilage, using a robotic system-termed here "3D histocutter"-which automatically sections tissue samples and serially acquires fluorescence microscopy images of each section. Tibiae dissected from C57Bl/6 mice, either naïve or OA-induced by surgical destabilization of the medial meniscus (DMM), were imaged using the 3D histocutter by exploiting tissue autofluorescence. Accuracy of 3D imaging was validated by ex vivo contrast-enhanced micro-CT and sensitivity to lesion detection compared with conventional histology. Reconstructions of tibiae obtained from 3D histocutter serial sections showed an excellent agreement with contrast-enhanced micro-CT reconstructions. Furthermore, osteoarthritic features, including articular cartilage loss and osteophytes, were also visualized. An in-house developed software allowed to automatically evaluate articular cartilage morphology, eliminating the subjectivity associated to semi-quantitative scoring and considerably increasing analysis throughput. The novelty of this methodology is, not only the increased throughput in imaging and evaluating mouse articular cartilage morphology starting from conventionally embedded samples, but also the ability to add the third dimension to conventional histomorphometry which might be useful to improve disease assessment in the model.

Diseased Region Detection of Longitudinal Knee Magnetic Resonance Imaging Data

Magnetic resonance imaging (MRI) has become an important imaging technique for quantifying the spatial location and magnitude/direction of longitudinal cartilage morphology changes in patients with osteoarthritis (OA). Although several analytical methods, such as subregion-based analysis, have been developed to refine and improve quantitative cartilage analyses, they can be suboptimal due to two major issues: the lack of spatial correspondence across subjects and time and the spatial heterogeneity of cartilage progression across subjects. The aim of this paper is to present a statistical method for longitudinal cartilage quantification in OA patients, while addressing these two issues. The 3D knee image data is preprocessed to establish spatial correspondence across subjects and/or time. Then, a Gaussian hidden Markov model (GHMM) is proposed to deal with the spatial heterogeneity of cartilage progression across both time and OA subjects. To estimate unknown parameters in GHMM, we employ a pseudo-likelihood function and optimize it by using an expectation-maximization (EM) algorithm. The proposed model can effectively detect diseased regions in each OA subject and present a localized analysis of longitudinal cartilage thickness within each latent subpopulation. Our GHMM integrates the strengths of two standard statistical methods including the local subregion-based analysis and the ordered value approach. We use simulation studies and the Pfizer longitudinal knee MRI dataset to evaluate the finite sample performance of GHMM in the quantification of longitudinal cartilage morphology changes. Our results indicate that GHMM significantly outperforms several standard analytical methods.

The repeatability of T2 relaxation time measurement of human knee articular cartilage

OBJECTIVES

To assess short- and long-term repeatability of T2 relaxation time measurements of the knee articular cartilage.

MATERIALS AND METHODS

The right knees of nine asymptomatic volunteers (age 30-38 years, five male, four female) were imaged at 1.5 T in three sessions 1 and 2 weeks apart. To observe short-term repeatability, the measurements were repeated three times within one of the three imaging sessions for each volunteer. T2 relaxation time was mapped using a multi-slice multi-echo spin echo sequence in axial and sagittal planes. Cartilage was manually segmented and repeatability, as measured by root-mean-square coefficient of variation (CVRMS) was evaluated both for the entire bulk cartilage of each joint surface in the slice and separately for each region of interest (ROI) at different topographical locations and separately for the superficial and deep half of each ROI.

RESULTS

For bulk T2, the long-term repeatability was 3.2, 5.4, and 3.7%, and the short-term reproducibility was 3.9, 3.9, and 3.4% for bulk femoral, tibial, and patellar cartilage, respectively. There were no significant differences between long-term and short-term repeatability in superficial or deep cartilage when comparing CVRMS values (p = 0.338 and 0.700, respectively). For individual ROIs, the repeatability varied between 2.5 and 22.2% depending on the topographical location.

CONCLUSION

The current results show mostly good repeatability. However, there were remarkable variations of T2 between bulk cartilage and different ROIs, bulk cartilage showing better repeatability. With careful patient positioning T2 can be accurately determined for different cartilage surfaces.

Automatic bone segmentation and bone-cartilage interface extraction for the shoulder joint from magnetic resonance images

We present a statistical shape model approach for automated segmentation of the proximal humerus and scapula with subsequent bone-cartilage interface (BCI) extraction from 3D magnetic resonance (MR) images of the shoulder region. Manual and automated bone segmentations from shoulder MR examinations from 25 healthy subjects acquired using steady-state free precession sequences were compared with the Dice similarity coefficient (DSC). The mean DSC scores between the manual and automated segmentations of the humerus and scapula bone volumes surrounding the BCI region were 0.926  ±  0.050 and 0.837  ±  0.059, respectively. The mean DSC values obtained for BCI extraction were 0.806  ±  0.133 for the humerus and 0.795  ±  0.117 for the scapula. The current model-based approach successfully provided automated bone segmentation and BCI extraction from MR images of the shoulder. In future work, this framework appears to provide a promising avenue for automated segmentation and quantitative analysis of cartilage in the glenohumeral joint.

Imaging research results from the osteoarthritis initiative (OAI): a review and lessons learned 10 years after start of enrolment

The Osteoarthritis Initiative (OAI) is a multicentre, prospective, observational, cohort study of knee osteoarthritis (OA) that began recruitment in 2004. The OAI provides public access to clinical and image data, enabling researchers to examine risk factors/predictors and the natural history of knee OA incidence and progression, and the qualification of imaging and other biomarkers. In this narrative review, we report imaging findings and lessons learned 10 years after enrolment has started. A literature search for full text articles published from the OAI was performed up to 31 December 2013 using Pubmed and the OAI web page. We summarise the rationale, design and imaging protocol of the OAI, and the history of OAI publications. We review studies from early partial, and later full OAI public data releases. The latter are structured by imaging method and tissue, reviewing radiography and then MRI findings on cartilage morphology, cartilage lesions and composition (T2), bone, meniscus, muscle and adipose tissue. Finally, analyses directly comparing findings from MRI and radiography are summarised. Ten years after the first participants were enrolled and first papers published, the OAI has become an invaluable resource to the OA research community. It has fuelled novel methodological approaches of analysing images, and has provided a wealth of information on OA pathophysiology. Continued collection and public release of long-term observations will help imaging measures to gain scientific and regulatory acceptance as 'prognostic' or 'efficacy of intervention' biomarkers, potentially enabling shorter and more efficient clinical trials that can test structure-modifying therapeutic interventions (NCT00080171).

Quantitative measurement of femoral condyle cartilage in the knee by MRI: validation study by multireaders

PURPOSE

To determine reproducibility of the femoral condyle cartilage volume (CV) in cross-sectional and longitudinal studies using various 3D imaging techniques at 1.5 T and 3 T.

MATERIALS AND METHODS

In 21 subjects with osteoarthritis, magnetic resonance imaging (MRI) including four different sequences (sagittal 3D fat suppressed spoiled gradient-echo [SPGR] at 1.5 T, fat suppressed fast low angle shot [FLASH] at 3 T, water-excitation dual echo steady state [DESS] at 3 T, and water-excitation multiecho data image combination [MEDIC] at 3 T) were acquired at baseline and ∼1 year later. The CV measured using semiautomated segmentation software by three readers was analyzed.

RESULTS

The mean of the interclass correlation coefficient between each reader from SPGR, FLASH, DESS, and MEDIC was 0.899, 0.948, 0.943, and 0.954, respectively. The mean CV (×10(4) mm(3) ) measured by each reader from SPGR/FLASH/DESS/MEDIC sequences was the following in this order: 1.34/1.52/1.50/1.35, 1.21/1.43/1.40/1.27, 1.22/1.37/1.36/1.22, and 1.17/1.36/1.35/1.21 by readers 1, 2, 3 (first analysis), and 3 (second analysis), respectively. There was no statistically significant difference in CV between any readers in any sequences. The CV measured on FLASH and DESS tended to be greater than that on SPGR or MEDIC.

CONCLUSION

Inter- and intraobserver reproducibility of cartilage segmentation using semiautomated software was validated. Although there was no statistical significance, there was a tendency of under- or overestimating CV by each sequence.

Registration of Magnetic Resonance Image Series for Knee Articular Cartilage Analysis: Data from the Osteoarthritis Initiative

OBJECTIVE

Although conventional radiography is used to assess osteoarthritis in a clinical setting, it has limitations, including an inability to stage early cartilage degeneration. There is a growing interest in using quantitative magnetic resonance imaging to identify degenerative changes in articular cartilage, including the large multicentered study, the Osteoarthritis Initiative (OAI). There is a demand for suitable image registration and segmentation software to complete this analysis. The objective of this study was to develop and validate the open source software, ImageK, that registers 3 T MRI T2 mapping and double echo steady state (DESS) knee MRI sequences acquired in the OAI protocol.

METHODS

A C++ library, the insight toolkit, was used to develop open source software to register DESS and T2 mapping image MRI sequences using Mattes's Multimodality Mutual information metric.

RESULTS

Registration was assessed using three separate methods. A checkerboard layout demonstrated acceptable visual alignment. Fiducial markers placed in cadaveric knees measured a registration error of 0.85 voxels. Measuring the local variation in Mattes's Mutual Information metric in the local area of the registered solution showed precision within 1 pixel. In this group, the registered solution required a transform of 56 voxels in translation and 1 degree of rotation.

CONCLUSION

The software we have developed, ImageK, provides free, open source image analysis software that registers DESS and T2 mapping sequences of knee articular cartilage within 1 voxel accuracy. This image registration software facilitates quantitative MRI analyses of knee articular cartilage.

Equivalence and precision of knee cartilage morphometry between different segmentation teams, cartilage regions, and MR acquisitions

OBJECTIVE

To compare precision and evaluate equivalence of femorotibial cartilage volume (VC) and mean cartilage thickness over total area of bone (ThCtAB.Me) from independent segmentation teams using identical Magnetic Resonance (MR) images from three series: sagittal 3D Dual Echo in the Steady State (DESS), coronal multi-planar reformat (DESS-MPR) of DESS and coronal 3D Fast Low Angle SHot (FLASH).

DESIGN

Nineteen subjects underwent test-retest MR imaging at 3 T. Four teams segmented the cartilage using prospectively defined plate regions and rules. Mixed models analysis of the pooled data were used to evaluate the effect of acquisition, team and plate on precision and Pearson correlations and mixed models were used to evaluate equivalence.

RESULTS

Segmentation team differences dominated measurement variability in most cartilage regions for all image series. Precision of VC and ThCtAB.Me differed significantly by team and cartilage plate, but not between FLASH and DESS. Mean values of VC and ThCtAB.Me differed by team (P < 0.05) for DESS, FLASH and DESS-MPR. FLASH VC was 4-6% larger than DESS in the medial tibia and lateral central femur, and FLASH ThCtAB.Me was 5-6% larger in the medial tibia, but 4-8% smaller in the medial central femur. Correlations between DESS and FLASH for VC and ThCtAB.Me were high (r = 0.90-0.97), except for DESS vs FLASH medial central femur ThCtAB.Me (r = 0.81-0.83).

CONCLUSIONS

Cartilage morphology metrics from different image contrasts had similar precision, were generally equivalent, and may be combined for cross-sectional analyses if potential systematic offsets are accounted for. Data from different teams should not be pooled unless equivalence is demonstrated for cartilage metrics of interest.

Relationship between cartilage volume using MRI and Kellgren-Lawrence radiographic score in knee osteoarthritis with and without meniscal tears

OBJECTIVE

The purpose of our study was to evaluate the correlation between the Kellgren-Lawrence (KL) radiographic score and the femoral and tibial cartilage volumes determined by MRI in patients with knee osteoarthritis. The effect of meniscal tears and extrusion on the cartilage volume was also examined.

SUBJECTS AND METHODS

Knee cartilage was evaluated by MRI in 74 patients (20 men and 54 women) who were categorized according to the KL score. Sagittal fat-suppressed 3D spoiled gradient-echo images were obtained to calculate the cartilage volume. The cartilage volume was determined for the lateral femoral cartilage, medial femoral cartilage, lateral tibial cartilage, and medial tibial cartilage. The femoral condylar bone volume was measured to adjust for bone size in each cartilage volume measurement.

RESULTS

After adjusting for age, sex, and femoral condylar bone volume, the cartilage volumes were significantly different between the grades in all compartments. Additionally, significant correlations were observed between the KL score and the adjusted cartilage volumes of lateral femoral cartilage and lateral tibial cartilage without a meniscal tear and between the KL score and the adjusted cartilage volume of medial femoral cartilage with and without a meniscal tear.

CONCLUSION

These findings showed a significant negative association between cartilage volume and the KL score. The cartilage volume of medial femoral cartilage may be more affected by the severity of osteoarthritis grade than the presence of a meniscal tear. In contrast, the cartilage volume in the lateral tibiofemoral compartment may be easily affected by the presence of a meniscal tear.

Quantification of cartilage loss in local regions of knee joints using semi-automated segmentation software: analysis of longitudinal data from the Osteoarthritis Initiative (OAI)

INTRODUCTION

Quantitative cartilage morphometry is a valuable tool to assess osteoarthritis (OA) progression. Current methodologies generally evaluate cartilage morphometry in a full or partial sub-region of the cartilage plates. This report describes the evaluation of a semi-automated cartilage segmentation software tool capable of quantifying cartilage loss in a local indexed region.

METHODS

We examined the baseline and 24-month follow-up MRI image sets of twenty-four subjects from the progression cohort of Osteoarthritis Initiative (OAI), using the Kellgren-Lawrence (KL) score of 3 at baseline as the inclusion criteria. A radiologist independently marked a single region of local thinning for each subject, and three additional readers, blinded to time point, segmented the cartilage using a semi-automated software method. Each baseline-24-month segmentation pair was then registered in 3D and the change in cartilage volume was measured.

RESULTS

After 3D registration, the change in cartilage volume was calculated in specified regions centered at the marked point, and for the entire medial compartment of femur. The responsiveness was quantified using the standardized response mean (SRM) values and the percentage of subjects that showed a loss in cartilage volume. The most responsive measure of change was SRM=-1.21, and was found for a region of 10mm from the indexed point.

DISCUSSION

The results suggest that measurement of cartilage loss in a local region is superior to larger areas and to the total plate. There also may be an optimal region size (10mm from an indexed point) in which to measure change. In principle, the method is substantially faster than segmenting entire plates or sub-regions.